RoadmapBench is a benchmark of 115 real version-upgrade tasks showing that even top AI coding agents succeed on fewer than 40% of long-horizon, multi-file software changes.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.SE 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
RoadmapBench: Evaluating Long-Horizon Agentic Software Development Across Version Upgrades
RoadmapBench is a benchmark of 115 real version-upgrade tasks showing that even top AI coding agents succeed on fewer than 40% of long-horizon, multi-file software changes.